TW201342919A - Data encoding and decoding - Google Patents

Abstract

Data coding apparatus in which a set of ordered data is encoded comprises an entropy encoder for encoding the ordered data, in which each data item is split into respective subsets of data and the respective subsets are encoded by first and second encoding systems so that for a predetermined quantity of encoded data generated in respect of a group of data items by the first encoding system, a variable quantity of zero or more data is generated in respect of that group of data by the second encoding system; and an output data stream assembler for generating an output data stream from the data encoded by the first and second encoding systems, the output data stream comprising successive packets of a predetermined quantity of data generated by the first encoding system followed, in a data stream order, by the zero or more data generated by the second encoding system in respect of the same data items as those encoded by the first encoding system.

Description

Data encoding and decoding [Related application]

The present invention claims that the entire contents of the applications are incorporated in the earlier application days of GB1119687.0 and GB1119180.6, which were filed on November 15, 2011 and November 7, 2011, respectively. reference.

The invention relates to data encoding and decoding.

The "prior art" provided herein generally represents the purpose of the disclosure herein. The current work of the inventor of the present invention, that is, the scope of the prior art and the description of the invention in the present invention may not be considered as prior art at the time of application, and thus does not represent or imply any prior art.

As an example of data encoding and decoding technology, there are several video data compression and decompression systems, which involve converting video data into frequency domain representation, quantizing frequency domain coefficients, and then applying some form of entropy coding to quantized coefficients.

In this context, entropy can be thought of as information content that represents a data symbol or series of symbols. The goal of entropy coding is to encode a series of data symbols in a lossless manner, (ideally) using a minimum number of encoded data bits, which are required to represent the data symbols of the series. In practice, the entropy coding is used to encode the quantized coefficients so that the encoded data (in terms of the number of bits) is smaller than the original The size of the data for the quantized coefficients. A more efficient entropy encoding process gives a smaller output data size for the same input data size.

One technique for entropy encoding video data is known as the so-called CABAC (Context Adaptation Binary Arithmetic Plus) technique. This is an example of a more generalized arithmetic plus code (AC) technique. In the illustrated embodiment, the quantized coefficients are partitioned into data relative to the position of the coefficient array representing coefficient values of certain sizes and their signs. Thus, for example, a "significance map" can mean a position in an array of coefficients at which the coefficients have a non-zero value. Other maps may also indicate that the material has a value of one or greater; or that the material has a value of two or greater.

In the basic example of CABAC encoders and decoders, the effect map is encoded as CABAC data, but some of the other maps are encoded as so-called bypass data (coded as CABAC, but with a fixed 50% probability context) model). The effect map and other graphs are all representatives of different individual attributes or ranges of values for the same initial data item. Thus, individual data items are grouped into individual subgroups and individual subgroups are encoded with a first (eg, CABAC) and a second (eg, bypass) encoding system.

Broadly speaking, the bypass data cannot be introduced into the same data stream as the CABAC coded material in its original form, but for any given output CABAC decoded data bit, the CABAC decoder has been encoding the particular data bit as the encoder is encoding. The meta-time encoder has already written more bits read from the data stream. In other words, in view of further reading of CABAC encoded data from the data stream, the CABAC decoder reads first, so it is generally considered impossible to introduce bypass data into the same CABAC data. Continuously encoded data stream.

The present invention provides a data plus code device, wherein a set of sorted data is encoded, comprising: an entropy encoder for encoding the sorted data, wherein each data item is divided into individual subgroups and individual subgroups are first and The second encoding system is encoded such that for a predetermined number of encoded data generated for a group of data items of the first encoding system, a variable number of zero or more data is associated with the data set for the second encoding system And generating an output stream combiner for generating an output data stream from the data encoded by the first and second encoding systems, the output data stream including a predetermined amount of data generated by the first encoding system A contiguous packet, in a data stream sequence, followed by zero or more data generated by a second encoding system having the same data item encoded for the first encoding system.

Embodiments of the present invention allow, for example, bypass information to be available at a predetermined location in the stream by splitting the CABAC data into packets of a predetermined length and then following each packet's bypass material (if any) ), which corresponds to the coefficient is encoded as CABAC data. Using this technique, the bypass data can be interpreted simultaneously with the CABAC data. Accordingly, embodiments of the present invention provide a method of segmenting CABAC streams such that bypass data can be placed in the stream in its original form to form a composite CABAC/bypass data stream and possibly with CABAC data (in Parallel interpretation when decoding.

In one example, in some systems, the so-called bypass data is encoded as a CABAC stream by encoding each bit as if it were a CABAC bit, but with a 50% fixed probability (context). This effect involves multiplication of the current range with the bypass bit when encoding, and requires recursive logic (or divider) to decode the multi-bit. Because the decoder does not have a way to know which bits are bypass data and which are CABAC data, the bypass data cannot be directly introduced into the bit stream.

However, by placing the bypass data in a particular location in the stream that is known to the encoder and decoder, it is possible not only to write the bypass data in the original binary form, but also to allow the multi-bit to be simply Read, while also allowing bypass bits to be decoded in parallel. This will allow bypass material (eg, symbols) to be decoded parallel to the CABAC data (eg, the effect map).

The following techniques can be assisted by arranging data into packets of a predetermined size to support parallel decoding.

Note that CABAC is only an example, and the present invention can be applied to other types of code inclusion including, but not limited to, general arithmetic plus code techniques.

Embodiments of the present invention may also provide a data encoder, wherein a buffer is used to accumulate data renormalized by the register, which represents a lower limit of the CABAC range, and is used to correlate the stored data into a group. If it has at least one predetermined amount of data; a detector for detecting whether all of the data in the group has a data value of 1 without a carry, and if so, for designating the group as a group of the first type; if not, Then the group is designated as a group of the second type; a buffer reader, if the subsequent storage group is of the second type, Reading, by the buffer, a group of the first type, and inserting the read group into the output data stream; and detecting, by the detector, detecting whether more than a predetermined number of the first type of groups appear in the buffer, if Yes, it is used to terminate and restart the encoding of the material.

Further individual aspects and characteristics of the present invention are defined in the scope of the accompanying patent application.

It is to be understood that the foregoing general description of the invention

DETAILED DESCRIPTION OF THE INVENTION Referring now to the drawings, Figures 1 through 4 are provided to provide a schematic representation of an apparatus or system utilizing a compression and/or decompression device in accordance with an embodiment of the present invention as described below.

All data compression and/or decompression devices are described in hardware as a general purpose data processing device, such as a general purpose computer software implementation, which is implemented as a programmable hardware, such as a specific application integrated circuit ( ASIC) or field programmable gate array (FPGA) or a combination thereof. When the embodiment is implemented in software and/or firmware, it will be appreciated that the software and/or firmware, and non-transitory machine can read data storage media in which such software and/or software are stored or provided. Firmware, and is considered to be an embodiment of the present invention.

FIG. 1 schematically illustrates an audio/video data transmission and reception system using video data compression and decompression.

The input audio/video signal 10 is supplied to a video data compression device 20 that compresses at least the video component of the video/audio signal 10 for transmission along a transmission path 30 such as a cable, fiber optic, wireless link or the like. . The compressed signal is processed by decompression device 40 to provide an output audio/video signal 50. For the return path, compression device 60 compresses the audio/video signals for transmission along transmission path 30 to decompression device 70.

The compression device 20 and the decompression device 70 can thus form a node of the transmission link. Decompression device 40 and decompression device 60 may form another node of the transmission link. Of course, for example, when the transmission link is unidirectional, only one node of these nodes requires a compression device, and the other node will only need to decompress the device.

Figure 2 is a schematic illustration of a video display system decompressed using video data. More specifically, a compressed audio/video signal 100 is processed by decompression device 110 to provide a decompressed signal that can be displayed on display 120. The decompression device 110 can be implemented as an integrated component of the display 120, such as within the same housing as the display device. Alternatively, decompression device 110 may be provided, for example, as a so-called set top box (STB), noting that the term "set" does not necessarily mean that the box needs to be positioned in any particular direction or position associated with display 120; it is simply The term used in the art refers to a device that can be connected to a display as a peripheral device.

Figure 3 shows an audio/video storage system using video data compression and decompression. The input audio/video signal 130 is supplied to a compression device 140 that generates a compressed signal for storage by the storage device 150, such as a disk device, a disk device, a tape device, a solid state A storage device, such as a semiconductor memory or other storage device. For playback, the compressed data is read by storage device 150 and transmitted to decompression device 160 for decompression to provide an output audio/video signal 170.

It will be appreciated that a compressed or encoded signal, and a storage medium storing the signal, are considered to be embodiments of the present invention.

Figure 4 shows schematically a camera compressed using video data. In FIG. 4, a video capture device 180, such as a charge coupled device (CCD) image sensor and associated control and readout electronics, generates a video signal that is transmitted to compression device 190. The microphone (or majority of microphones) 200 produces an audio signal that is transmitted to the compression device 190. Compression device 190 generates a compressed audio/video signal 210 that is stored and/or transmitted (generally represented by exemplary stage 220).

The techniques described below are primarily related to video data compression. It will be appreciated that many existing techniques can be used with video data compression techniques described below for audio data compression to produce compressed audio/video signals. Therefore, a separate discussion of audio data compression will not be provided. It will be appreciated that the data rate for video material, especially broadcast quality video data, is generally much higher than the data rate of the audio material (whether compressed or uncompressed). Therefore, it can be appreciated that the uncompressed audio material will be accompanied by compressed video data to form a compressed audio/video signal. It is further understood that although this example (shown in Figures 1 through 4) pertains to audio/video data, the techniques described below can be useful in systems that simply handle (ie, compress, decompress, store, display, and / or transmit) video material. That is, the Embodiments can be applied to video data compression without necessarily having any associated audio data processing.

Figure 5 provides a schematic overview of a video data compression and decompression device.

The input video signal 300 of the continuous video is supplied to the adder 310 and to the image predictor 320. Image predictor 320 will be described in more detail below with reference to FIG. In effect, adder 310 performs a subtraction (negative addition) operation that receives the input video signal 300 at the "+" input and the output of image predictor 320 at the "-" input such that the input image is subtracted from the predicted image. This result is produced by a so-called residual image signal 330 representing the difference between the actual and projected images.

The reason why a residual image signal is generated is as follows. The data plus code technique is described as the technique that will be applied to the residual image signal will tend to work more efficiently when there is less "energy" in the image being encoded. Herein, the term "valid" means that a smaller amount of encoded material is generated for a particular image quality level, and we want (and think "effective") to produce data that is as small as possible. The "energy" mentioned in the residual image is related to the amount of information contained in the residual image. If the projected image is the same as the real image, then the difference between the two (ie, the residual image) will contain zero information (zero energy) and will be easily encoded into a small amount of encoded data. In general, if the estimation process can reasonably perform well, it is expected that the residual image material will contain less information (less energy) than the input image, and at the same time, it will be more easily encoded as a small amount of coded material.

The residual image data 330 is supplied to a conversion unit 340 which produces a discrete cosine transform (DCT) representation of the residual image data. DCT technology It is known per se and will not be detailed here. However, the technical aspects used in the device of the present invention will be described in more detail below, more specifically regarding the selection of different block data to which DCT operations are applied. These will be discussed below with reference to Figures 7-12.

The output of the conversion unit 340, that is, a set of DCT coefficients for each conversion block of the image material, is supplied to the quantizer 350. Various quantization techniques are known in the field of video data compression, ranging from simple multiplication to quantization of scaling factors to the application of complex look-up tables under the control of quantization parameters. There are two general goals. First, the quantization process reduces the number of possible values of the converted material. Second, the quantization process can increase the likelihood that the value of the transformed data is zero. Both of these can make the entropy encoding process as described below more efficient to produce a small amount of compressed video material.

The data scanning process is applied by a scanning unit 360. The purpose of the scanning process is to reorder the quantized conversion data to obtain as many non-zero quantized conversion coefficients as possible, and thus of course obtain as many zero value coefficients as possible. These characteristics may allow so-called scan length overweight or similar techniques to be more effectively applied. Therefore, the scanning process involves selecting coefficients by the quantized conversion data, more specifically, selecting coefficients by the coefficient block corresponding to the image data blocks that have been converted and quantized, so that (a) all coefficients are selected. Once as part of the scan, and (b) the scan tends to provide the ones you want to reorder. The technique for selecting the scanning order will be explained as follows. An exemplary scanning sequence that tends to give useful results is the so-called zigzag scanning sequence.

The scan coefficients are then passed to an entropy encoder (EE) 370. once again, Various types of entropy coding can be used. The two examples which will be described below are variants of the so-called CABAC (Context Adaptation Binary Arithmetic Plus) system and variants of the so-called CAVLC (Context Adaptation Variable Length Plus) system. In common language, CABAC is believed to provide better efficiency, and in some studies it has been shown to provide a 10%-20% reduction in the number of encoded output data compared to CAVLC comparable image quality. However, CAVLC is considered to exhibit a much lower complexity than CABAC (indicated by the implementation). The CABAC technique will be discussed with reference to Figure 17 below, and the CAVLC technique will be discussed with reference to Figures 18 and 19 below.

Note that the scan processing and the entropy encoding processing are shown as separate processing, but may in fact be combined or processed together. That is to say, the read-in entropy encoder of the data can occur in the scanning order. Corresponding considerations apply to each of the inverse processes described below.

The output of the entropy encoder 370 and the additional data (described above and/or below), for example, to define the manner in which the predictor 320 produces the predicted image, provides a compressed output video signal 380.

However, because the operation of the predictor 320 itself depends on the decompressed version of the compressed output data, a return path is also provided.

The reason for this feature is as follows. In the appropriate level of decompression processing (as described below), a decompressed version of the residual data is generated. This decompressed residual data must be added to the estimated image to produce an output image (because the original residual data is the difference between the input image and the predicted image). For this process to be comparable on the compression side to the decompression side, the estimated image produced by the predictor 320 should be the same during the compression process and the decompression process. Of course, when decompressing, The device does not have to use the original input image, but only the decompressed image. Thus, at the time of compression, the predictor 320 estimates (at least for inter-image encoding) an uncompressed version of the compressed image.

The entropy encoding process performed for the entropy encoder 370 is considered "lossless", that is, it can be reversed to obtain the same material that was first supplied to the entropy encoder 370. Therefore, the return path can be implemented before the entropy coding stage. Indeed, the scanning process performed for scanning unit 360 is also considered lossless, but in the present embodiment, return path 390 is the input from quantizer 350 to the input of complementary inverse quantizer 420.

In the common language, the entropy decoder 410, the reverse scan unit 400, the inverse quantizer 420, and the inverse conversion unit 430 provide individual reversal functions of the entropy encoder 370, the scanning unit 360, the quantizer 350, and the conversion unit 340. Now, the discussion will continue through the compression process; the process of decompressing the input compressed video signal will be discussed separately as follows.

In the compression process, the scan coefficients are passed to the inverse quantizer 420 by the quantizer 350, which returns the path 390, which performs the reversal operation of the scanning unit 360. The inverse quantization and reverse conversion processing is performed by units 420, 430 to generate a compressed-decompressed residual image signal 440.

Image signal 440 is added to the output of predictor 320 at adder 450 to produce reconstructed output image 460. This is formed into one of the input of the image predictor 320 and will be discussed below.

Now, the process of applying to the received compressed video signal 470, which is supplied to the entropy decoder 410 and supplied to the inverse scan unit from the adder 450 before being applied to the output of the image predictor 320, will be discussed. 400. Reverse the chain of the quantizer 420 and the reverse conversion unit 430. In the direct language, the output 460 of adder 450 forms an output decompressed video signal 480. In fact, further filtering can also be applied before the signal is output.

Figure 6 is a schematic representation of the generation of an estimated image, particularly with respect to the operation of image predictor 320.

There are two basic models of estimation: the so-called intra-image estimation and so-called inter-inage, or motion compensation (MC) estimation.

The intra-image prediction is based on the prediction of the content of the block image in the same image. This corresponds to the so-called I-frame coding within other video compression techniques. Different from the I-frame coding, in which the entire image is intra-coded, in this embodiment, the selection between intra-encoding and inter-encoding can be completed on a block-to-block basis, but In other embodiments of the invention, the selection can still be done on an image-to-image basis.

The motion compensation estimate utilizes motion information that attempts to define the source of the image detail encoded in the current image, which is in another adjacent or nearby image. Thus, in an ideal example, the content of the block image data in the predicted image can be simply encoded as a reference (motion vector) pointing to the corresponding block at the same or slightly different position in the adjacent image.

Returning to Figure 6, as shown in the two image estimation configurations (corresponding to intra-image and inter-image prediction), the result is under the control of the mode signal 510, The multiplexer 500 is selected to provide an estimated image block supply to the adders 310 and 450. The selection is based on which selection gives the lowest "energy" (as discussed above, which can be considered to be the information content that needs to be encoded), and the selection is sent to the encoder within the encoded output data stream. In this context, image energy can be detected by performing an attempt to subtract the area of the two versions of the predicted image from the input image; square the pixel values of the difference image; sum the squared value; and indicate which of the two versions Causes the lowest mean square value of the difference image for the image area.

In the inner coding system, the actual estimate is based on the image block received as part of the signal 460, that is, the estimate is done according to the coded-decoded image block to complete in the decompression device. The same estimate. However, the data can be derived from the input video signal 300 by the internal mode selection 520 to control the operation of the intra-image predictor 530.

For the inter-image estimation, the motion compensation (MC) predictor 540 uses, for example, motion information of the motion vector derived from the input video signal 300 of the motion evaluator 550. These motion vectors are applied to the processed version of the reconstructed image 460 by the motion compensation predictor 540 to produce an inter-image predicted block.

The process applied to signal 460 will now be described. First, the signal is filtered by a filtering unit 560. This involves applying a "deblocking" filter to remove or at least tend to reduce the effects of the blocks performed by the conversion unit 340 for the primary and subsequent operations. At the same time, an adaptive loop filter is applied using coefficients derived by processing the reconstructed signal 460 and the input video signal 300. Adaptable ring filters are a type of filter that uses known techniques Techniques are applied to adapt the filter coefficients to the data to be filtered. That is, the filter coefficients can be varied depending on various factors. The data defining which filter coefficients are used is included as part of the encoded output data stream.

The filtered output from filter unit 560 in effect forms an output video signal 480. It is also buffered in one or more image stores 570; the storage of the continuous images is a requirement for motion compensation estimation processing, particularly the generation of motion vectors. To save storage requirements, the images stored in image storage 570 can be maintained in a compressed form and then decompressed to generate motion vectors. Any known compression/decompression system can be used for this particular purpose. The stored image is transmitted to an interpolation filter 580 which produces a higher resolution version of the stored image; in this example, an intermediate sampling (subsampling) is generated such that the internal output of the interpolation filter 580 is output. The resolution of the interpolated image is 8 times the resolution (in each dimension) stored in image storage 570. The interpolated image is transmitted as an input to the action evaluator 550 and also to the motion compensated predictor 540.

In an embodiment of the present invention, another selection stage is provided, which is to multiply the data value of the input video signal by a multiplier 600 by a factor of four (actually shifting only the data value to the left two-digit element) and use The divider or right shifter 610 applies a corresponding division operation (right shift two bits) at the output of the device. Therefore, the left shift and the right shift change data are simply used as internal operations of the device. This approach can provide higher computational accuracy within the device as the impact of any data pick-up errors is reduced.

The manner in which the image is divided into compression processing will now be described. At the base layer In the second, the image to be compressed is considered to be an array of sampling blocks. For the purposes of this discussion, the largest of these considerations is the so-called maximum plus unit (LCU) 700 (Fig. 7), which represents a square array of 64x64 samples. Here, the discussion is about luminance sampling. Depending on the chromaticity mode of 4:4:4, 4:2:2, 4:2:0 or 4:4:4:4 (GBR plus key data), there will be corresponding colors corresponding to the chrominance blocks. Different numbers for sampling.

The three basic types of blocks will be described: an overweight unit, an estimation unit, and a conversion unit. In general, the recursive subdivision of the LCU allows the input picture to be segmented such that the block size and block plus parameters (e.g., predictive or residual plus mode) can be set based on the particular characteristics of the image being encoded.

The LCU can be subdivided into so-called oversampling units (CUs). The coded unit is always square and has a size between 8x8 samples and the full size of the LCU 700. The coded units may be arranged in a tree structure of a type such that the first subdivision may occur as shown in Figure 8, giving a 32x32 sampled adder unit 710; subsequent subdivisions may occur on a selection basis to give A portion of the 16x16 sampled partial adder unit 720 (FIG. 9) and some 8x8 sampled possible adder units 730 (FIG. 10). Overall, this process can provide the content of the CU block to accommodate the overweight tree structure, each of which can be as large as LCU or as small as 8 x 8 samples. The encoding of the output video material takes place according to the structure of the adding unit.

Figure 11 shows schematically an array of prediction units (PUs). The estimation unit is a basic unit for carrying information about the image estimation process, or in other words, additional information added to the entropy coded residual image data. The device of Figure 5 forms an output video signal. In general, the estimation unit is not limited to those whose shape is square. They may take any other shape, especially a rectangle forming one half of one of the square plus code units, as long as the coded unit is larger than the minimum (8 x 8) size. The goal is to allow the boundaries of neighboring prediction units (as close as possible) to match the boundaries of the real objects in the picture so that different prediction parameters can be applied to different real objects. Each of the coded units may include one or more estimation units.

Figure 12 shows schematically an array of conversion units (TUs). The conversion unit is the basic unit of conversion and quantization processing. The conversion unit is always square and can be sized from 4 x 4 to 32 x 32 samples. Each of the coded units may include one or more conversion units. The abbreviation SDIP-P in Fig. 12 simplifies the so-called short-range intra-predictive segmentation. In this configuration, only one-dimensional conversion is used, so a 4×N block is transmitted to the converter through N-conversion, and the input data is based on previously decoded neighboring blocks and previous decoded adjacent lines in the current SDIP-P. .

This internal estimation process will now be discussed as follows. In general, the inner estimate involves the generation of an estimate of the current block (estimation unit) from previously encoded and decoded samples within the same image. Figure 13 shows a partial encoded image 800. Here, the image is encoded from the upper left to the lower right on the basis of the LCU. An exemplary LCU coded halfway through the processing of the entire image is shown as block 810. The shaded area 820 above and to the left of block 810 is encoded. An intra-image estimate of the content of block 810 can use any shaded area 820, but cannot be used without an under-shaded area.

As discussed above, block 810 represents the LCU; for intra-image prediction For the purpose of rationality, this can be further subdivided into a smaller set of prediction units. The exemplary prediction unit 830 is displayed within the LCU 810.

The intra-image estimates take into account the sampling above and/or to the left of the current LCU 810. The sample of the desired sample to be estimated may be located at a different location or direction relative to the current prediction unit within the LCU 810. In order to determine which direction is applicable to the current prediction unit, the prediction results for each candidate direction are compared to see which candidate direction results are closest to the corresponding block of the input image. The candidate direction that gives the closest result is selected as the estimated direction for the prediction unit.

The picture can also be encoded on a "slice" basis. In one example, the slices are LCUs of horizontally adjacent groups. In general, however, the entire residual image may form a slice, or a slice may be a single LCU, or a slice may be a column of LCUs or the like. Slices can provide some flexibility in giving errors because they are encoded as separate units. The encoder and decoder states are completely reset at the slice boundary. For example, the inner estimate is not executed across the slice boundary, and the slice boundary is processed as the image boundary for this purpose.

Figure 14 illustrates the display of a set of possible (candidate) prediction directions. The entire set of 34 candidate direction systems can be used for estimation units of 8x8, 16x16 or 32x32 samples. A special example of estimated cell sizes for 4x4 and 64x64 samples has a reduced candidate direction for the group (17 candidate directions and 5 candidate directions, respectively). These directions are determined by the horizontal and vertical displacements relative to the current block position, but are encoded as an estimated "mode", one of which is shown in Figure 15. Note that the so-called DC mode represents a simple arithmetic average of the surrounding upper and left hand samples.

FIG. 16 schematically shows a zigzag scan which is one of the scan patterns that can be applied by the scanning unit 360. In FIG. 16, the pattern is shown as an exemplary block of 8x8 DCT coefficients, the DC coefficient is positioned at the upper left position 840 of the block, and the horizontal and vertical spatial frequencies are increased to the right of the upper left position 840. The coefficient of distance is expressed.

Note that in some embodiments, the coefficients may be scanned in reverse order (bottom right to top left, using the ranking of Figure 16). At the same time, it should be noted that in some embodiments, the scan may pass through several (eg, between one and three) uppermost horizontal columns from left to right before performing the tortuosity of the residual coefficients.

Figure 17 schematically illustrates the operation of a CABAC entropy coder.

The CABAC encoder operates in binary data, ie, the data is represented only by two symbols of 0 and 1. The encoder uses a so-called context modeling process that selects a "context" or likelihood model for the next data based on the previously encoded material. The selection of the context is performed in a deterministic manner such that the same decision from the previously decoded material can be performed at the decoder without further data (specifying the context) being added to the encoded data stream transmitted to the decoder.

Referring to Figure 17, if the input data to be encoded is not in binary form, it can be passed to binary converter 900; if it is already in binary form, converter 900 is skipped (schematic switch 910). In the present embodiment, the conversion to the binary form is actually performed or represented by a representation of the quantized DCT coefficient data as a series of binary "maps", as will be further described below.

Binary data can then be "normal" and "bypass" for both processing paths (It is illustrated as a separate path, but may be implemented in one embodiment of the invention discussed below for the same processing stage, but with slightly different parameters). The bypass path uses a so-called bypass coder 920, which does not necessarily use contextualization of the same form as the normal path. In some examples of CABAC overwriting, if there is a need for a fast processing of a particular batch of data, then this bypass path can be selected, but in this embodiment, the two characteristics of the so-called "bypass" data are noticed. : First, the bypass data is handled by the CABAC encoder (950, 960), using only a fixed context model representing 50% probability; and second, the bypass data is related to certain categories of data, a specific case is Coefficient symbol data. Alternatively, the normal path is selected for the schematic switches 930, 940. This involves the processing of the data for the context modeler 950 followed by a plus code engine 960.

The entropy encoder shown in Fig. 17 encodes a block of data (i.e., data corresponding to a coefficient block having respect to a residual image block) to a single value if the block is formed by the entire zero value material. For each box that does not fall into this category, that is, at least some of the blocks with non-zero data, a "effect map" is prepared. The effect map indicates that for each position in a data block to be encoded, if the corresponding coefficient in the block is non-zero. The value data for the binary form is itself CABAC coded. The use of the effect map assists in compression because there is no data to be encoded as a coefficient having a size for a coefficient having a number of its effect maps representing zero. At the same time, the effect map may contain special codes to represent the final non-zero coefficients in the block such that all final high frequency/tailing zero coefficients may omit the encoding. In the coded bit stream, the value map is defined as the value specified in the effect map. The data of the value of the non-zero coefficient is followed.

The other stages of the map data are prepared and coded. An example is a map that defines whether a binary value (1 = yes, 0 = no) has been specified as "non-zero" in the map position of the effect map, and actually has a value of "one." The other map indicates whether the coefficient data whose map position value map is indicated as "non-zero" actually has a value of "2". Another figure shows that the effect map already shows the map locations where the coefficient data is "non-zero" and whether the data has a value greater than two. For data that is again designated as "non-zero", another figure represents the symbol of the data value (using a predetermined binary representation, such as 1 for +, 0 for - and of course other ways).

In an embodiment of the invention, the effect map and other maps are configured in a predetermined manner to the CABAC encoder or to the bypass encoder and are all representatives of different individual attributes or ranges of values for the same initial data item. In an example, at least the effect is a CABAC code and at least a portion of the residual map (eg, symbol data) is bypass coded. Thus, each data item is segmented into individual subgroup data and individual subgroups are encoded by a first (eg, CABAC) and a second (eg, bypass) encoding system. The nature of the data and the nature of CABAC and bypass coding are such that for a predetermined amount of CABAC coded data, a variable number of zero or more bypass data is generated relative to the same initial data item. Thus, for example, if the quantized reordering DCT data contains values that are substantially all zero, there may be no bypass data or a small amount of bypass data, since the bypass data only has a care effect graph indicating that the value is non-zero. Figure location. In another example, in quantitative reordering DCT data with very high value coefficients, a significant amount of bypass may be generated. material.

In the embodiment of the present invention, the effect value map and other graphs are generated by the quantized DCT coefficients of the scanning unit 360 and subjected to a zigzag scan process before being subjected to CABAC encoding (or according to the internal prediction mode, by zigzag, horizontal Scan and scan the selected scan processing).

In general, CABAC coding involves estimating the context, or a probability model of the next bit of the element encoded according to other previously encoded data. If the next bit is the same as the "almost identical" bit indicated by the likelihood model, then the encoding of the information for the "next bit in accordance with the likelihood model" can be encoded more efficiently. The "next bit does not match the probability model" is less efficient, so the derivation of the context data is important for the good operation of the encoder. The term "adaptive" means that the context or likelihood model is adapted or changed at the time of encoding to provide a good match for the next data (eg, uncoded).

Using a simple analogy, in the written English, the letter "U" is quite uncommon. But in the letter position after the letter "Q", it is really common. Therefore, the probability that the probability of setting "U" may be a very low value, but when the current letter is "Q", the probability model of "U" as the next letter may be set to a high probability value.

In this configuration, CABAC encoding is used, and for at least the effect graph and the graph representation whether the non-zero value is one or two. Bypass processing - in these embodiments is the same as CABAC coding, but in fact, the likelihood model is fixed at 1 and 0 equal (0.5:0.5) likelihood distribution and is used for at least symbol data and graph representation of values. >2. For this indicated as >2 The location of these data, a separate separation of the data encoding can be used to encode the actual value of the data. This can include Golomb-Rice coding techniques.

The CABAC context modeling and coding process is described in more detail in TD4: 2011-10-28 201 x (E) "Working Draft 4 for Efficient Video Plus", JCTVC-F803_d5, Draft ISO/IEC 23008-HEVC.

Fig. 18 illustrates a CAVLC entropy encoding process.

Regarding the CABAC discussed above, the entropy encoding process shown in FIG. 18 follows the operation of the scanning unit 360. It has been noted that the non-zero coefficients in the converted and scanned residual data are often in the order of ±1. The CAVLC adder represents the number of high frequency ±1 coefficients in a variable called "Trail 1" (T1s). For these non-zero coefficients, the additive coefficients are improved by using different (context adaptive) variable length adder tables.

Referring to Figure 18, a first step 1000 produces a value of "coeff_token" to encode the total number of non-zero coefficients and the number of trailing ones. At step 1010, each trailing one of the symbol bits is encoded in a reverse scan order. At step 1020, each residual non-zero coefficient is encoded as a "hierarchy" variable, thus defining the sign and size of these coefficients. At step 1030, the variable total_zeros is used to add the total number of zeros before the last non-zero coefficient. Finally, at step 1040, the variable run_before is used to add the number of consecutive zeros before each non-zero coefficient in the reverse scan order. The output of the variables defined above forms the encoded data.

As described above, the predetermined scanning order for the scanning operation performed by the scanning unit 360 is a zigzag scanning and is illustrated in FIG. In other configurations of four squares using intra-code encoding, it can depend on the image prediction direction. (Fig. 15) and the conversion unit (TU) size, and choose between zigzag scanning, horizontal scanning and vertical scanning.

The CABAC process as discussed above will be described in more detail.

At least as the CABAC used in the proposed HEVC system involves deriving a "context" or likelihood model associated with the encoding of a bit. The context defined for the context variable or CV then affects how the bit is encoded. In general, if the next meta-system is defined to have the same value as the expected more likelihood value for the CV, it would be advantageous to reduce the number of output bits required to define the data bit.

The encoding process involves mapping a bit to a position within a range of code values. The range of code values is schematically illustrated in Figure 19A as a series of adjacent integers extending from the lower limit m_low to the upper limit m_high. The difference between these two limits is m_range, where m_range=m_high-m_low. In the basic CABAC system, the m_range is limited to between 256 and 512 by various techniques discussed below. M_low can be any value. It can start, for example, from zero, but can also be changed as part of the encoding process described.

The range of code values m_range is the relevant context variable defined for the boundary 1100 subdivided into two ranges: boundary = m_low + (CV * m_range)

Therefore, the context variable subdivides the total range into two ranges or sub-parts, one for the value with respect to zero (or the next data bit) and the other for the one (or the next data bit) The value of ). The subdivision of the range represents the CV for the next encoded bit as the two-dimensional value. Produce the assumed probability. Thus, if there is a sub-range for the value zero that is less than half of the total range, then this simplifies the possibility that zero is considered to be lower than 1 at the next symbol.

There are various probabilities of which to define which way the range applies to possible data bit values. In one example, the lower region range (ie, from m_low to boundary) is traditionally defined as a data bit value associated with zero.

The encoder and decoder maintain records of which data bit values are low or (often referred to as "least likelihood symbols" or LPS). CV represents LPS such that CV always represents a value between 0 and 0.5.

The next bit (the current input bit) is now mapped or assigned to the code value within the appropriate sub-range within the range m_range divided by the boundary. This is determined by the encoder and decoder using the techniques detailed below. If the next bit is 0, then a particular code value representing the position within the sub-range from m_low to the boundary is assigned to the bit. If the next bit is 1, a particular code value in the sub-range from the boundary 1100 to m_high is assigned to the bit.

The lower limit m_low and the range m_range are then redefined to modify the set of code values in relation to the specified code and the size of the selected sub-range. If the coded bit is zero, then m_low is unchanged but m_range is redefined to be equal to m_range*CV. If the coded bit is 1, then m_low is moved to the boundary position (m_low+(CV*m_range)) and m_range is redefined as the difference between the boundary and m_high (ie, (1-CV)*m_range)).

These alternative methods are illustrated in Figures 19B and 19C.

In Figure 19B, the data bit is 1 so the m_low is moved up to the previous boundary position. This provides the code value of the revised set for use in the next metacode order. Note that in some embodiments, the value of the CV is changed for the next bit-encoding, at least in part to the value of the just-encoded bit. This is why the technology is called the "adaptation" context. The revised value of the CV is used to generate a new boundary 1100'.

In Fig. 19C, the zero value is encoded, so m_low remains unchanged, but m_high is moved to the previous boundary position. The value m_range is redefined as the new value of m_high-m_low. In this example, this has caused m_range to fall below its minimum allowable value (eg, 256). When this result is detected, the value m_range is multiplied by 2, ie, shifted 1 bit to the left, as many times as necessary to reply m_range to 256 to 512 of the desired range. In other words, the set of code values is continuously increased in size until it has at least a predetermined minimum size. This example is shown in Figure 19D, which represents twice the range of Figure 19C to meet the required limitations. The new boundary 1100" is derived from the next value of the CV and the revised m_range.

As long as the range must be multiplied by two in this way, this process is often referred to as "renormalization", an output bit is generated (becomes an output encoded data bit), and one bit is used for each renormalization level.

In this way, the spacing m_range is continuously modified and renormalized depending on the CV value (which can be reproduced at the decoder) and the adaptation of the encoded bit string. After encoding a series of bits, the resulting spacing and the number of renormalization levels uniquely define the encoded bit stream. Know the decoding of this final spacing The device will primarily rebuild the encoded data. However, built-in mathematics shows that it does not actually need to define the spacing of the decoder, but only a position within the spacing. This is the purpose of assigning a code value that is maintained at the encoder and transmitted to the decoder (as the final part of the data stream) at the end of encoding the data.

The context variable CV is defined as having 64 possible states that continuously indicate different likelihoods from a lower limit of CV = 63 (eg, 1%) to a probability of being at CV = 0 to 50%.

The CV is changed from one bit to the next bit according to various known factors, and the factors may vary depending on the block size of the data to be encoded. In some examples, the state of adjacent and previous image blocks can be considered.

The specified code value is generated by a table that defines the code value of the new coded bit that should be specified in the position or position group in the relevant sub-range for each possible CV value and each possible value of m_range bits 6 and 7 ( Note that bit 9 of m_range is always 1 because of the limitation in m_range size).

Fig. 20 schematically illustrates a CABAC encoder using the above technique.

The CV is initialized (at the first CV) or modified (at the subsequent CV) by the CV derivation unit 1120. The code generator 1130 subdivides the current m_range according to the CV and uses the above table to generate the specified material code within the appropriate sub_range. Range reset unit 1140 resets m_range to a range of selected sub-ranges. If necessary, the normalizer 1150 renormalizes m_range and outputs an output bit for each of these renormalization operations. As mentioned earlier, at the end of the process, the specified code value is also output.

In the decoder shown in Fig. 21, the CV is initialized (when the first CV) or modified (at the subsequent CV) by the CV derivation unit 1220, which operates in the same manner as the unit 1120 in the encoder. The code application unit 1230 subdivides the current m_range according to the CV and detects which range the data code is placed. Range reset unit 1240 resets m_range to the range of the selected secondary range. If necessary, normalizer 1250 renormalizes m_range in response to receiving the data bit.

In summary, the technique of the present invention allows CABAC data (i.e., data using context variables) to be written to a bit stream in a 16-bit (in this example) fixed size packet, which is referred to as a 'CABAC packet'. After each 'CABAC packet', a corresponding 'bypass packet' is written to the bit stream.

A 'bypass packet' (which is variable in size) contains any bypass bits that are attached to the CABAC data, which can be decoded using only the bits contained in the previous 'CABAC packet'; this bypass data It is inserted directly into the stream.

To generate a CABAC packet as the main stream, the encoder can be arranged to track how many bits the decoder has read after each renormalization process. The encoder can start counting at a number equal to the number of bits (9 in this embodiment) that the decoder was initially read.

Some other background information will be provided.

Referring now to Figures 22 and 23, as described above, an entropy coder forming part of a video encoding device includes a first encoding system (e.g., an arithmetic plus code encoding system such as CABAC encoder 2400) and a second encoding system (example) For example, the bypass encoder 2410) is arranged such that a particular data character or value is encoded by the CABAC encoder or the bypass encoder but not both to the final output data stream. In an embodiment of the invention, the data values transmitted to the CABAC encoder and the bypass encoder are sequential data values of individual subgroups, which are segmented from the initial input data (in this example, reordered quantized DCT data). Or export, representing a different map of the set of "maps" produced by the input data.

The CABAC encoder and the bypass encoder are separately configured in the schematic representation of FIG. This may be the case in practice, but in another possibility, a single CABAC encoder 2420 as illustrated in FIG. 23 is used as the CABAC encoder 2400 and the bypass encoder 2410 of FIG. The encoder 2420 operates under the control of the mode selection signal 2430 to operate in an adaptive context model (as described above) in the mode of the CABAC encoder 2400, and in the mode of the bypass encoder 2410 to fix 50 % contingency context model operation.

The third possible combination is that both substantially identical CABAC encoders can operate in parallel (similar to the parallel configuration of Figure 22) with the difference that the CABAC encoder operating as bypass encoder 2410 has its context model fixed at 50% Probability context model.

The output of the CABAC encoding process and the bypass encoding process may be stored (at least temporarily) in the individual buffers 2440, 2450. In FIG. 23, a switch or demultiplexer 2460, under the control of mode signal 2430, distributes CABAC encoded data to buffer 2450 and bypass encoded data to buffer 2440.

An alternate configuration using a single buffer will be described below with reference to FIG.

24 and 25 schematically illustrate an example of an entropy decoder forming portion of the video decoding device. Referring to Figure 24, the individual buffers 2510, 2500 transmit data to the CABAC decoder 2530 and the bypass decoder 2520, which are configured such that a particular encoded data word or value is decoded by a CABAC decoder or a bypass decoder, But not both. The decoded data is reordered by logic 2540 into the appropriate order for subsequent decoding stages.

The schematic representation of the CABAC decoder and the bypass decoder in Fig. 24 are separately configured. In fact, this may be the case, but in another possibility, as shown in FIG. 25, a single CABAC decoder 2550 is used as both the CABAC decoder 2530 and the bypass decoder 2520 of FIG. The decoder 2550 operates under the control of the mode selection signal 2560 to operate with an adaptive context model (as described above) when in the mode of the CABAC decoder 2530, and a fixed 50% operation when bypassing the mode of the encoder 2520. Probability context model.

As before, the third may combine the two, in that the two substantially identical CABAC decoders can operate in parallel (similar to the parallel configuration of Figure 24), with the difference being that the CABAC decoder operating as the bypass decoder 2520 has its context model fixed. In the 50% likelihood context model.

In the example of FIG. 25, a switch or multiplexer 2570 operates under the control of mode signal 2560 to properly distribute CABAC encoded data to decoder 2550 by buffer 2500 or buffer 2510.

The embodiments of the present invention will be further described in detail below, CABAC coding Material and bypass coded data can be multiplexed into a single data stream. More details of the data streaming will be explained below, but in this phase, note that in this configuration, the input buffers 2500, 2510 and/or the output buffers 2440, 2450 (in this example) may be a single individual buffer. Replaced by 2580. Thus, in a decoder configuration, the two input buffers can be replaced by a single buffer, and in an encoder configuration, the two output buffers can be replaced by a single buffer. In Figure 26, the buffer is schematically illustrated to include vertical lines of delimited data bits or characters that are intended to be representative of the data content (as shown) extending across the buffer in the landscape.

The buffer 2580 and its associated read and write control configuration can therefore also be considered to be used to generate an output stream of data encoded by the first (eg, CABAC) and second (eg, bypass) encoding systems. An example of an output data combiner.

Two buffer indicators 2590, 2600 are shown. When the encoder outputs a buffer, these representative data are written to the indicator, indicating where the next data bit is written in the buffer. In the example of a decoder input buffer, these representations indicate a data read indicator of the location in the buffer where the next data bit was read. In an embodiment of the invention, the indicator 2590 is related to reading or writing CABAC encoded data and the indicator 2600 is related to reading or writing bypass encoded data. The effectiveness of these indicators and their associated locations will be discussed below.

In the basic example of a CABAC encoder and decoder, the code bypass data (encoded as CABAC but with an inherent 50% likelihood context model) cannot introduce the same data stream as the CABAC decoded data in its original form, for any given output. CABAC decoding data bit, the The CABAC decoder has been written to read more bits from the data stream than when the encoder is encoding the particular data bit. In other words, the CABAC decoder reads the front to read further CABAC encoded data from the data stream, and therefore does not consider it possible to introduce bypass data to the same continuous encoded data stream as the CABAC data. This difference (the number of forward reads by the decoder) can be referred to as "decoder offset."

In other words, when the decoder is processing a binary value, it already has some bits for the lower binary values in the scratchpad, which is called a "value".

However, if one way to find out that the bypass material can be used in its original form, most of the bits of the bypass data can be read immediately with relatively little logic or processing overhead. Embodiments of the present invention do allow for the bypass material to be taken at a predetermined location in the stream. Using the techniques described, the bypass data can be read simultaneously with the CABAC data. Accordingly, embodiments of the present invention provide a method for segmenting a CABAC stream such that bypass data can be placed in the stream in its original form to form a composite CABAC/bypass data stream and possibly parallel to CABAC data. Read (at the time of decoding).

The basis of this technology is to arrange CABAC data streams (without bypass data) as packets. Herein, the packet is represented as a set of adjacent coded CABAC data bits having a predetermined length (number of bits), wherein the term "predetermined" implies that the length of the CABAC data packet is, for example, (a) determined in advance, and (b) is an encoder. Determining and transmitting the remaining encoded data to the decoder in a stream, or (c) by means of a decoder and an encoder known Pre-encoding/decoding data is exported.

After each CABAC packet is written to the output data stream, the bypass data corresponding to the encoding coefficient contained in the packet is written (in raw form) to the composite output data stream.

Thus, this configuration provides an example of output data stream generation comprising a contiguous packet of a predetermined amount of data generated by a first coding system (e.g., CABAC), and in the data stream sequence, followed by correlation The same data item encoded by the first encoding system is zero or more data generated by the second encoding system (eg, bypass).

The encoder tracking decoder will read how many bits after each decoding to determine the amount of bypass material after the next packet.

The decoder can load the CABAC packet into a buffer (e.g., buffer 2510) and read the bypass material directly from the stream. Alternatively, the CABAC and bypass data can be read by a common buffer or stream using separate indicators (described with reference to Figure 26). In this way, it is possible that most of the bypass bits can be read at once, and the CABAC and bypass data can be read in parallel, making it possible (possibly) to be encoded as a single with respect to CABAC data and bypass data in the system using common arithmetic plus code processing. The system of data streaming increases the data traffic of the system.

Therefore, by dividing the CABAC data into packets in a direct manner, it is possible to combine the original bypass data with the CABAC data, allowing most of the bits to be read at once, and/or simultaneously reading the bypass and CABAC data, allowing parallelism and improvement. Flux.

In an embodiment of the invention, a fixed size 16-bit system is used CABAC packet. Note that this fixed length is the number of output data produced; of course, the nature of the CABAC encoding process means that the 16 bits in the CABAC packet can of course represent a variable amount of input data. The 16-bit length is greater than the CABAC range (preset 9 bits) used in embodiments of the present invention. Also note that other packet lengths can also be selected.

The CABAC data that initializes the register "value" is placed in the first packet, followed by the CABAC data to be renormalized after the first decoding, the second decoding, and so on. Each CABAC envelope is followed by the original bypass data, and is used to renormalize all of the coefficients of the CABAC data system that are contained within the CABAC packet.

The bypass data for coefficients with renormalized data execution through the end of a particular CABAC packet is placed after the next CABAC packet.

This process is schematically illustrated in Figure 27, which shows a portion of the packetized data stream. The CABAC packet 2610 is shown as an unshaded square and the bypass packet 2620 is shown as a shaded square with data order from left to right. Padding zero 2630 (see below) is shown to be related to the last CABAC packet. Thus, the packetized data stream of Figure 27 provides an example of video data comprising a predetermined number of consecutive packets of ordered data generated by the first encoding system, in the data stream sequence, followed by a first encoding associated with Zero or more data generated by a second encoding system of the same data item encoded by the system, the first and second encoding systems being arranged to entropy encode the input video data of the individual subgroups such that the first encoding system is A predetermined amount of encoded data generated by a group of data items, related to the data generated by the group as the second encoding system, produces a variable amount of zero or more data.

When the video data is encoded on a slice-by-slice basis (one of the slices is a subgroup of a picture, so that the decoding is self-contained, or the other part is independent of the picture, and the independent coding is a separate part), The last packet for a particular slice can be smaller than the expected 16 bits if the CABAC data for that slice does not end at the packet boundary (that is, if the total CABAC data for the entire slice is not a multiple of 16 bits). In these environments, the last packet will be less than 16 bits in length. At this time, the last CABAC packet is padded by the data stream combiner, for example, padding the data stream combiner to zero to its expected size, writing the padding data to the buffer and reading it out as the last packet. Part of it. The padding data is not read because the decoder will decode the flag at the end of the slice before it is taken. In this example, the maximum amount of wear is equal to one bit less than the packet size.

However, in order to reduce the amount of wear, if the final CABAC packet has no associated bypass material (zero bypass data), the final packet may be allowed to be shorter than expected - or in other words, the padding data is not used. This package contains what can be called a "short package."

In an embodiment of the invention, to encode data in the form of a packet, the encoder maintains a buffer (e.g., buffer 2580) in which CABAC and bypass data can be written. At initialization, the CABAC write indicator 2590 is set to zero and the bypass write indicator 2600 is set equal to the size of the first CABAC packet (eg, 16 bits). Therefore, the first bit of the CABAC data will be written to the beginning of the buffer, and the first bit of the bypass material (if any) associated with the CABAC packet will be Write to a position shifted by 16 bits by the first CABAC write position. This indicator configuration is illustrated in more detail in Figure 28.

Whenever the encoder detects that the decoder has read 16 bits, the bypass indicator proceeds through the next CABAC packet (ie, 16 bits are advanced by the end of the bypass data for the current packet). Whenever the encoder fills in the CABAC packet, the CABAC write indicator is forwarded to the stream through the next bypass packet and CABAC and bypass data.

To encode a CABAC packet as a main stream, an output buffer can be used, which can be written to multiple locations.

Two write metrics are used to index this buffer, the first to indicate where the CABAC data is written (starting at zero) and the second to indicate where the bypass data is written (starting at 16).

Whenever the encoder detects that the decoder has read 16 bits, the position of the bypass indicator is noted and incremented by 16. When the encoder completes the current 'CABAC Packet' write, the CABAC indicator is set equal to the attention previous position of the Bypass indicator.

In this way, each indicator 'skips' other information. After any given renormalization, the difference between the total number of bits read by the decoder and the total number of bits written by the encoder is already greater than the size of a packet.

Therefore, it is necessary to store most of the previous bypass indicator locations. The required number of indicators can be constrained by the maximum difference and is the reason for limiting the final position.

Non-word aligned writes can be performed on the buffer by caching the bytes surrounding the target area within a single write cycle.

The following steps are virtual codes that describe the current encoding process. The interpretation of some notes is done in parentheses when necessary.

Initialization: set ptr_CABAC = 0 (CABAC write indicator) ptr_bypass = 16 (bypass write indicator) ptr_bypass_old = undefined decodet_bits_read = 9 (CABAC_RANGE_BITS)

Step 1: Encode the N CABAC bit: If the N bit fits into the current packet: write the bit ptr_CABAC+=N in ptr_CABAC (replace ptr_CABAC with ptr_CABAC+N) otherwise write the bit in ptr_CABAC until the end of the packet ptr_CABAC=ptr_bypass_old ptr_CABAC writes the remaining bits (N') ptr_CABAC+=N'decoder_bits_read+=N if decoder_bits_read>=16 decoder_bits_read-=16 ptr_bypass_old=ptr_bypass ptr_bypass+=16

Step 2: Code N bypass bit: write bit ptr_bypass+=N in ptr_bypass

Step 3: Repeat steps 1 and 2 as needed.

In coding design, a termination symbol is required at the end of a set of data, and it will be appreciated that a variable number of CABAC bits processed in the last packet should still allow for this termination symbol. The termination symbol can be, for example, a bit having a predetermined value of 0 or 1, representing a network abstraction layer (NAL) data. The end of the stream.

The use of '[...]' denotes a CABAC packet previously described, 'C' denotes a CABAC bit and 'R' denotes a terminating symbol, and the final (short) packet may then, for example, contain the bit order of the type [CCCCCCCCCCCCCCCR]. In this example, there are 15 CABAC bits and a terminating bit.

Note that in the HEVC or AVC system, in the embodiment of the present invention, the last bit of the CABAC stream itself can be replaced by this termination symbol, thus saving one bit. In this example, the final bit of the CABAC stream is replaced with the same value as the termination symbol of the encoding design (eg, a value of one). Obviously, if the final value of this bit is already 1, there is no change in effect. If the final value of the bit is 0, it has the effect of adding 1 to the entire value. However, for a symbol range of 2 and a value for the bottom of the symbol range at the end of the encoding, adding 1 will not move the entire value beyond the plus code spacing, and the data stream remains valid. Thus, for HEVC or AVC encoding, typically, the last bit of the CABAC stream can be set equal to the termination symbol while remaining within the same plus code spacing. It will be appreciated that any over-coded design can utilize this technique to similarly allow for bit changes in the last bit in the stream.

Therefore, in this embodiment, if the CABAC stream ends in the middle of the packet and no bypass data can be followed, the stream is terminated as a short packet as described above, and the last bit of the short packet will be Read as a proper termination symbol and can be decoded correctly (eg, in the HM4.0 decoding design). Using the above example, the resulting short packet will read [CCCCCCCCCCCCCCR], containing 14 CABAC bits and a terminating bit to replace the last 15th CABAC bit.

However, if the CABAC stream terminates in the middle of the packet and there is bypass data to follow, the final packet is not truncated to coincide with the end of the CABAC bit, and conversely provided with padding bits. However, I still want to terminate the symbol.

In this example, in one embodiment of the invention, one or more padding bits (typically all padding bits in this embodiment) are also used with the terminator in order to maintain consistency between the encoder and the decoder. The same value. Use 'B' for the bypass bit and the resulting order will be of type [CCCCCCCCCCRRRRRR]BBBBR. Here, 10 CABAC bits are followed by 6 padding bits having the same value or sign as the data terminating bit. After the CABAC packet, the bypass bit is then followed by the termination symbol again.

Again, in HEVC, AVC or similar tolerance designs, or the final bit of the CABAC bit can be replaced again with a terminating symbol, with additional padding bits also using the same value as the terminating symbol. At this point, the resulting order will be of type [CCCCCCCCCRRRRRRR]BBBBR. Here, the nine CABAC bits are followed by a terminating bit to replace the final 10th CABAC bit, followed by six padding bits having the same value as the terminating bit R. Thus, there is a total of 7-bit columns, in this example, having the value of the termination symbol R. After the CABAC packet, the bypass bit is followed again by the termination symbol.

In either case, in this way, the termination symbol is still set at the end of the CABAC data and also at the end of the entire bit stream.

Finally, when the final required CABAC bit (ie, based on the coding design described above, with or without the last actual CABAC bit) is exactly matched to the 16th bit of the final CABAC packet, the decoder will automatically continue to view Where an expected packet will be and thus the termination symbol can be placed at that location. Thus, for example, the order in this example includes [CCCCCCCCCCCCCCCC]R, and [CCCCCCCCCCCCCCCC]BBBBR.

Separately or additionally, it is also possible to terminate the CABAC stream earlier to insert different data (eg, IPCM lossless code). As a result, the CABAC stream can be terminated again as a short packet, but when there is bypass data again, the final packet can again contain padding bits. In this example, one or more of the bits that again save the padding bits may be replaced by the corresponding first bit (the value taken) at the beginning of the insertion of the data (subsequent stream). Using 'D' to indicate the difference data, the resulting packet will then contain a sequence type [CCCCCCCDDDDDDDDD] BBBBR or (refer to HEVC or AVC example above) [CCCCCCCRDDDDDDDD] BBBBR.

The pending bits will be considered with reference to Figure 29. As part of this consideration, in an embodiment of the invention, the encoder maintains a list of indicators in which the bypass packet is completed. The length of the list depends on the decoder offset described above. In general: Decoder offset=CABAC_RANGE_BITS (default 9) + number of final bits

In general, the decoder offset is theoretically the last constraint, however, its proposed mechanism is provided to limit the pending bits, allowing the encoder to maintain a smaller number of metrics.

The required number of metrics is given by: rounding (maximum decoder offset / CABAC packet size)

The pending bit is a possible problem for the arithmetic decoder. When the encoder knows that the encoded data is in the range of 0.49 to 0.51 (for example, in decimal, not in general binary) - the decoder does not know whether to output 0.4 ( The first effect map) or 0.5. The encoder must delay outputting information until it knows what to do. For each delay decision, the encoder must write one bit less than needed to the stream, so the decoder offset increases. Finally, the problem is solved and the encoder can write the missing value and reduce the decoder offset back to 9. However, it must write all missing values to the correct location in the buffer, so it must be remembered where it can be written to the stream buffer.

Therefore, it must maintain the number of tracking locations as a function of decoder offset and packet size. If the offset is <=16, it can be learned to keep track of only normal write metrics and older write metrics, but if the offset is >16, the extra older write locations need to be tracked; if offset > 32, then the two additional older entry locations need to be recorded, and so on.

Another aspect of the pending bit will be discussed. In order to output the stream in the correct order, the "bypass packet" must be stored by the encoder until its corresponding "CABAC packet" is generated.

Because the encoder can delay writing to the bit for a long time, or even the entire stream, since these bits are pending, the number of "bypass packets" that must be buffered can be infinite.

In order to limit the number of 'bypass packets' that must be buffered, for example using a stream termination, it is useful to limit the number of pending bits.

This method allows the restricted bits to be limited to a fixed number without having to be checked after each renormalization. To accomplish this, when m_low appears in the encoder, the decoder tracks m_low.

The bits renormalized by m_low are accumulated in a buffer. If at least a portion of the bits have accumulated in this buffer after renormalization, they are considered to form a 'group'. In an embodiment of the invention, this minimum group size may be 15 bits. Therefore, the maximum group size will be equal to 22 (14 + 8 (8 bits is the maximum possible renormalization)).

In the encoder, the first group is stored in a buffer. If the subsequent group is not pending, the group currently in the buffer is flushed into the stream along with any pending bits. The new group is then stored in this buffer.

If all of the bits in a group are 1 and there is no carry, the group is considered pending. The encoder and decoder each maintain a count of the number of pending groups. The encoder also maintains a count of the sum of the size of the pending group so that when no unresolved group is encountered, it knows how many pending bits need to be flushed.

If the number of unresolved groups is greater than or equal to a defined limit, the stream is terminated. This determines that the next group will not be pending, allowing the cumulative undecided group to be flushed.

With this method, the maximum possible count of pending bits is equal to the limit number of pending groups multiplied by the maximum group size.

Thus, embodiments of the present invention may provide a buffer for accumulating data renormalized by m_low (a register indicating the lower limit of the CABAC range), and if the group has at least one predetermined amount of data, Store data into a group; a detector for detecting whether all of the data in the group has a data value of 1 without a carry, and if so, for designating the group as a group of the first type; if not, the group is designated as the second type a buffer reader for reading a first type of group from the buffer, if the subsequent storage group is of a second type, and inserting the read group into the output data stream; a detector for detecting There are more than a predetermined number of groups of the first type in the buffer, and if so, to terminate and resume encoding the material.

The operation of this decoder will now be described.

Expressed in the usual language, a CABAC packet-based main stream can be decoded using the shift register as the buffer. Each time data is read by the buffer as a decoder or control logic provided in the decoder as a buffer reader, and the remaining bits are shifted to fill the space occupied by the read bit. And new data from the stream is added to the end of the scratchpad.

The CABAC data is read into the m_value (as an input buffer) by the front end of the shift register, both at the beginning of decoding and after each renormalization.

The bypass data is read by the position indicated by the bypass index. This index is initially set to 16 (the beginning of the first 'bypass packet') and the number of read CABAC data decrement read bits per read.

In a parallel system, CABAC and bypass read-and-shift occur simultaneously. For example, the decoders of decoders 2520 and 2530 can act as an entropy decoder to decode the individual subgroups of data read by the buffer.

At the end of the 'CABAC packet' (ie, when CABAC reads the transfer) The road index), the bypass index is incremented by 16, skipping it through the next 'CABAC packet'. By inferring the reading by a position equal to the bypass index +16 and the bypass index itself, this can be done in a parallel system without stalling.

The data in the 'bypass package' is only attached to the CABAC data in the previous 'CABAC packet'. By CABAC reading will pass the current 'CABAC packet' end, this ensures that all data in the relevant 'Bypass Packet' will be read and therefore, the end of the current 'CABAC Packet' is directly adjacent to the next start .

This allows CABAC data to be safely read through the end of the current 'CABAC packet' as if it were a continuous stream.

The size of the shift register is calculated to be 67 bits by the maximum possible number of read bypass bits required to read from the farthest possible position and assuming that the coefficient has a maximum size of 32768 (16-bit symbols) ( (8-1+16) {farth bypass read index} + 44 {maximum possible bypass read}).

The decoder can maintain a buffer or shift register (e.g., buffer 2580 of the type shown in Figure 26) to allow both types of data (CABAC and bypass) to be operated simultaneously.

The minimum size of the buffer = (CABAC_RANGE_BITS-2) + PACKET_SIZE + MAX_BYPASS_LENGTH; for example, using a set of previously proposed parameters, the minimum size = (9-2) + 16 + 44 = 67.

The bypass data is read from the location indicated by the bypass read indicator 2600. An example of a buffer in use is shown schematically in Figure 30.

The decoder infers the bit after the bypass reading indicator to determine how many bits are to be read, if:

A) No bypass bit is required (coefficient size = 0)

B) only one symbol bit is needed (coefficient size = 1 or 2)

C) Symbol bits and escape codes (all other sizes) are required.

In addition, the decoder also infers decoding (bypass indicator +16) (if valid) (see below).

Referring to FIG. 30, after each CABAC renormalization, the CABAC data is read into the register "value" from the front of the buffer (the front end is displayed on the left side of FIG. 30), and all other bits are shifted to the left. Read indicator 2600 with bypass. In other words, in this example, the material read by the buffer has the effect of removing the data by the buffer. (depending on the size determined by the CABAC decoding process) The bypass data is read by the position indicated by the bypass indicator 2600, and all bits are shifted again to the right of the bypass read indicator to further buffer down. (left of the device).

The amount of new data equal to the sum of the two shifts is added to the end of the buffer by the encoded data stream (i.e., the right hand end as shown in Figure 30). In other words, the buffer is refilled after these read operations.

(When CABAC reads the indicator through the Bypass indicator) At the end of the CABAC packet, the Bypass Index is skipped to the next location through the next CABAC packet.

The bypass packet must be empty at this point (because the bypass bit is in the packet corresponding to the entire coefficient contained in the CABAC packet) so that the CABAC indicator can be safely read through the end of the packet as if it were It is a continuous stream.

Once the inference evaluation bit is in (bypass reading indicator +16), in this case In this case it is possible to ensure that the next bypass packet is immediately jumped without any additional processing delay.

This processing is schematically illustrated in Fig. 31. Note that the CABAC packet shown to the left is a schematic illustration of the current unread content of the current packet, that is, it only shows the remainder of the current packet.

In general, the inference indicator is always higher than the main indicator of 16 bits: the idea is that if there is a need to access the data in the main indicator (because it is at the beginning of the new packet), the decoder will have already decoded at the beginning of the next bypass packet. The first bypass data. In other words, this assists in ensuring that the bypass material has been interpreted.

The following virtual code, using the same labeling as the previous virtual code, describes the decoding process and can be read in conjunction with Figure 32, which illustrates four processing stages, as shown in Figure 32 as a continuous line, noting that embodiments of the present invention In this, four stages can be completed in each decoding clock cycle: Initialization: Fill the register "Value" with (CABAC_RANGE_BITS) from stream Fill shift register from stream index_bypass=(16-CABAC_RANGE_BITS)(set bypass read pointer)

Step 1a (parallel to step 1b): Decode CABAC data:Determine symbol ranges Compare the register "value" with symbol ranges to determine symbol(magnitude)Determine number of renormalisation bits for CABAC data(Nc)value<<=Nc

Step 1b (parallel to step 1a): Decode bypass data: Decode sign bit Sn from position index_bypass; Decode escape data En from position(index_bypass+1); Nbn=(escape length+1)Decode sign bit Ss from position(index_bypass+ 16); Decode escape data Es from position(index_bypass+17); Nbs=(escape length+1)

Step 1c (after steps 1a and 1b): if(Nc>index_bypass)- New packet encountered-use speculative path escape=Es;sign=Ss;Nb_escape=Nbs;Index_bypass+=16 Else escape=En;sign-Sn;Nb_escape =Nbn If(magnitude>2)output=(magnitude+escape)* sign;Nb=Nb_escape Else if(magnitude>0)output=(magnitude * sign);Nb=1 Else output=0;Nb=0

Step 2: shift and refill: Shift register<<=Nc index_bypass-=Nc Shift register[index_bypass to end]<<=Nb Read(Nc+Nb)bits from stream into last bits of shift register

Dividing CABAC data into packets can be particularly beneficial for intra-image modes. In the inter-image mode, the conversion coefficient data is often sparse in the matrix, and some CU/LCUs may not contain coefficients at all. Because inter-mode motion vectors may often provide better predictions than are available in intra-image modes, therefore, quantized residual data can often be small/non-existent.

Higher density data in the internal mode places higher demands on the encoder/decoder flow and, therefore, a more desirable parallelism.

These methods can have great advantages for the implementation method.

CABAC and bypass data can be read simultaneously, allowing parallel decoding.

Multi-bit bypass data can be read at the same time. This allows all bypass data for coefficients to be decoded in one stage.

Only a small decoder buffer is needed. A 16-bit buffer is required for a 16-bit packet size and a 9 CABAC_RANGE_BITS.

These techniques can be adapted to work in a multi-factor system per cycle.

These technologies can increase traffic.

Accordingly, the foregoing discussion discloses and describes only exemplary embodiments of the invention. The invention may be embodied in other specific forms without departing from the spirit and scope of the invention. Therefore, the disclosure of the present invention is intended to be illustrative, and not to limit the scope of the invention. The disclosure of the immediately-available changes included in the teachings herein defines a part of the scope of the aforementioned patent application, such that the no-invention subject matter belongs to the public.

10‧‧‧Optical/Video Signals

20‧‧‧Video compression equipment

30‧‧‧Transmission routing

40‧‧‧Decompression equipment

50‧‧‧Audio/Video Signal

60‧‧‧Compression equipment

70‧‧‧Decompression equipment

100‧‧‧Compressed audio/video signals

110‧‧‧Decompression equipment

120‧‧‧ display

130‧‧‧Audio/Video Signal

140‧‧‧Compression equipment

150‧‧‧Storage device

160‧‧‧Decompression equipment

170‧‧‧Optical/Video Signals

180‧‧‧Image capture device

190‧‧‧Compression equipment

200‧‧‧ microphone

210‧‧‧Compressed audio/video signals

220‧‧‧Storage and / or transmission

300‧‧‧Input video signal

310‧‧‧Adder

320‧‧‧Image Estimator

330‧‧‧Residual image signal

340‧‧‧Conversion unit

350‧‧‧Quantifier

360‧‧‧Scan unit

370‧‧‧Entropy encoder

380‧‧‧Compressed output video signal

390‧‧‧Return path

400‧‧‧Reverse Scanning Unit

410‧‧‧ Entropy decoder

420‧‧‧ inverse vectorizer

430‧‧‧Reverse conversion unit

440‧‧‧Residual image signal

450‧‧‧Adder

460‧‧‧ output image

470‧‧‧Compressed video signal

480‧‧‧Uncompressed video signal

500‧‧‧Multiplexer

510‧‧‧ mode signal

520‧‧‧Mode selector

530‧‧Intra-image predictor

540‧‧‧motion compensation predictor

550‧‧‧Action Estimator

560‧‧‧Filter unit

570‧‧‧Image storage

580‧‧‧Interpolation filter

600‧‧‧multiplier

610‧‧‧ divider

700‧‧‧Maximum overweight unit

710‧‧‧Plus unit

720‧‧‧Plus unit

730‧‧‧Plus unit

800‧‧‧Partial coded image

810‧‧‧ square

820‧‧‧Shaded area

830‧‧‧ Estimation unit

840‧‧‧Upper position

900‧‧‧ converter

910‧‧‧ switch

920‧‧‧ Bypass adder

930‧‧‧ switch

940‧‧‧ switch

950‧‧‧Context Modeler

960‧‧‧Plus engine

1120‧‧‧CV derivation unit

1130‧‧ ‧ code generator

1140‧‧‧ Range reset unit

1150‧‧‧Normalizer

1220‧‧‧CV derivation unit

1230‧‧‧ code application unit

1240‧‧‧ Range reset unit

1250‧‧‧Normalizer

2400‧‧‧CABAC encoder

2410‧‧‧bypass encoder

2420‧‧‧CABAC encoder

2430‧‧‧ mode selection signal

2440‧‧‧buffer

2450‧‧‧buffer

2460‧‧‧Solution multiplexer

2500‧‧‧buffer

2510‧‧‧buffer

2520‧‧‧Bypass decoder

2530‧‧‧CABAC decoder

2540‧‧‧Logic

2550‧‧‧CABAC decoder

2560‧‧‧ mode selection signal

2570‧‧‧Switch

2580‧‧‧buffer

2590‧‧‧Buffer indicator

2600‧‧‧Buffer indicator

A more complete understanding of the present disclosure and many of its advantages will be better understood by referring to the following detailed description of the embodiments of the present invention in conjunction with the drawings, wherein: FIG. 1 is a schematic diagram showing audio/video (compressed and decompressed using video data) A/V) data transmission and reception system; Figure 2 shows a video display system decompressed using video data; Figure 3 shows an audio/video storage system compressed and decompressed using video data; Figure 4 is a schematic view showing a video camera using video data compression; Figure 5 is a schematic diagram showing a video data compression and decompression device; Figure 6 is a schematic view showing the generation of an estimated image; Figure 7 is a schematic view showing a maximum coding unit (LCU); Four plus code units (CU); Figures 9 and 10 schematically show that the code unit of Figure 8 is subdivided into smaller code units; Figure 11 shows an array of prediction units (PU); Figure 12 shows an array of conversion units ( Figure 13 shows a partially encoded image; Figure 14 shows a set of possible prediction directions; Figure 15 shows a set of prediction modes; Figure 16 shows a zigzag scan; Figure 17 shows a CABAC entropy; 18 schematically shows CAVLC entropy coding processing; FIGS. 19A to 19D schematically show CABAC encoding and decoding operation modes; FIG. 20 schematically shows a CABAC encoder; FIG. 21 schematically shows a CABAC decoder; and FIG. 22 schematically shows CABAC with separate bypass encoders Encoder; Figure 23 shows schematically the encoder, actuated as CABAC encoder and bypass encoder; Figure 24 shows schematically CABAC decoding with separate bypass decoder Figure 25 shows a decoder that is actuated as a CABAC decoder and a bypass decoder; Figure 26 shows a common buffer; Figure 27 shows a packetized data stream; Figures 28 and 29 show data write indicators. Use; Figures 30 and 31 schematically show the use of data reading indicators; and Figure 32 shows the stages in the CABAC and bypass decoding processing operations.

2590‧‧‧Buffer indicator

2600‧‧‧Buffer indicator

Claims (18)

A data adding device, wherein the encoding has a set of sequential data, the device comprises: an entropy encoder for encoding the sequential data, wherein each data item is divided into individual subgroup data, and the individual subgroups are first and Encoding by the second encoding system such that for a predetermined number of encoded data generated for a group of data items for the first encoding system, a variable amount of zero or more data is associated with the second encoding of the group of data And an output data stream combiner for generating an output data stream from the data encoded by the first and second encoding systems, the output data stream being included in the first encoding system A contiguous packet of a predetermined number of data, the data stream sequence being followed by the same data item encoded for the first coding system for the zero or more data generated by the second coding system. The device of claim 1, wherein the first encoding system is an arithmetic plus code encoding system, and the second encoding system is a bypass encoding system. The apparatus of claim 2, wherein the first coding system is a context adaptive binary arithmetic plus code (CABAC) coding system. The apparatus of any one of claims 1 to 3, wherein the set of sequential data represents one or more images. The apparatus of any one of claims 1 to 4, wherein the set of sequential data is encoded as an independent encoded portion, the data stream The combiner is configured to generate a data stream associated with a portion, and when the final packet is less than a predetermined amount of data, the padding data is added to the final packet encoded for the first encoding system in relation to the sequential data . The device of claim 5, wherein one or more bits of the padding material have the same value as the data termination symbol. The device of claim 5, wherein the one or more individual bits of the padding data take the value of the corresponding individual bit at the beginning of the subsequent data stream. The apparatus of claim 5, wherein the data stream combiner is configured to not add padding data to the final package if zero data is generated for a corresponding coefficient encoded by the second encoding system. The device of any of the preceding claims, further comprising a frequency domain converter for generating frequency domain coefficients and sorting the coefficients for coding according to the coding order, depending on respective portions of the input data signal . A data overwriting method, wherein the encoding has a set of sequential data, the adding method comprising the steps of: dividing each data item into different sub-group data; entropy encoding the respective sub-groups by the first and second encoding systems, so that The first encoding system is associated with a predetermined amount of encoded data generated by a group of data items, a variable amount of zero or more data being associated with the group of data for the second encoding system; and And the data encoded by the second encoding system generates an output data stream, the output data stream including a continuous packet of a predetermined amount of data generated by the first encoding system, in a data stream sequence, followed by Zero or more data generated for the second encoding system associated with the same data item encoded for the first encoding system is followed. A video material encoded by the encoding method described in claim 10 of the patent application. A video material comprising a contiguous packet of a predetermined amount of data generated by a first encoding system, in a data stream sequence followed by a second encoding system associated with the same data encoded for the first encoding system Zero or more data generated by the item, the first and second encoding systems being arranged to entropy encode the respective sub-groups of the input video data such that a predetermined amount is generated for a group of data items associated with the first encoding system The encoded data, a variable number of zero or more data associated with the group of data for the second encoding system. A data carrier storing video material as described in claim 11 or 12. A video decoding device for decoding data as described in claim 11 or 12, the device comprising: an input buffer for storing the encoded data; and a buffer reader for reading by each packet The first and second sets of respective data; and an entropy decoder for decoding the first and second subgroups to generate ordered decoded data. A data decoding method for decoding video data as described in claim 11 or 12, the method comprising the steps of: storing the encoded data in an input buffer; Reading, by each packet in the input buffer, respective data of the first and second subgroups; and entropy decoding the first and second subgroups to generate sorted decoded data. A computer software that, when executed by a computer, causes the computer to perform the method of claim 10 or 15. A non-transitory machine readable storage medium having stored thereon a computer software as described in claim 16 of the patent application. A video data capture transmission and/or storage device comprising the apparatus of any one of claims 1 to 9 and 13.